Argoverse: 3D Tracking and Forecasting with Rich Maps

November 2020

tl;dr: Map data helps with 3D tracking and motion forecasting.

Overall impression

This dataset is one of a kind and focuses on tracking and motion forecasting (of other traffic agents other than ego car). The only other dataset that focuses on this is Lyft 3D dataset.

The original of the ego is center of the real axel. All lidar data is provided in the ego vehicle coordinate system.

Given the multimodal nature of the motion forecasting problem,. It might not be fair to evaluate against a single GT.

The paper used a CV-based lidar detection method. This is more traditional method used since the DARPA challenge.

Key ideas

Technical details